A Comparative Analysis of Community Detection Algorithms on Artificial Networks.

Area of Science:

• Network Science
• Computer Science
• Data Analysis

Background:

• Community detection algorithms aim to reveal mesoscopic properties in complex networks.
• Evaluating algorithm performance (accuracy, computing time) on real-world networks is challenging due to size and objective community definition limitations.

Purpose of the Study:

• To systematically evaluate and compare eight state-of-the-art community detection algorithms.
• To provide guidelines for selecting the most suitable algorithm based on network characteristics.
• To identify limitations of specific algorithms concerning macroscopic network properties.

Main Methods:

• Utilized the Lancichinetti-Fortunato-Radicchi benchmark graph for algorithm testing.
• Quantified algorithm accuracy using complementary measures and assessed computing time.
• Analyzed the influence of network size on algorithm performance and computational efficiency.

Main Results:

• Developed guidelines for selecting community detection algorithms based on observable network properties.
• Identified the mixing parameter as a key indicator for algorithm reliability ranges.
• Demonstrated the dependency of algorithm predicting power and computing time on network size.

Conclusions:

• The study offers practical techniques for choosing the most appropriate community detection algorithm.
• Provides insights into algorithm reliability based on the mixing parameter and network size.
• Enables informed selection of community detection methods for complex network analysis.